Squid doesn't share memory or disk cache at the moment. It won't share/slice filedescriptors the way you want them to. I could probably write a "unified" logging hack so multiple squid processes log to the same file via a single helper that handles multiple pipes or something, one from each Squid. There's no atomic "append a line" IO method in UNIX so doing it that way won't work. You could try hacking things up to lock/unlock the file for each logfile write but I have no idea what the impact would be. Adrian 2009/8/20 Joel Ebrahimi <jebrahimi@xxxxxxxxx>: > Hi, > > Im trying to build a high performance squid. The performance actually > comes from the hardware without changes to the code base. I am a > beginning user of squid so I figured I would ask the list for the > best/different way of setting up this configuration. > > The architecture set up is like this: There are 12 cpu cores that each > run an instance of squid, each of these 12 cores has access to the same > disk space but not the same memory, each is its own instance of an OS > and they can communicate on an internal network, there is a network > processor that slices up sessions and can hand them off to any one of > the 12 cores that is available, there is a single conf file and a single > logging directory. > > The current problem I can see with this set up is that each of the 12 > instances of squid acts individually, therefore any one of them could > try to access the same log file at the same time. Im not sure what > impact this could cause with overwriting data. > > I actually have it set up this way now and it works well though it's a > very small test environment and Im concerned issues may only pop up in > larger environments when accessing the logs is very frequent. > > I was looking through some online materials and I saw there are other > mechanisms for log formatting. The ones that I thought may be of use > here are either the daemon or udp. There is actually a 13th core in the > system that is used for management. I was wondering if setting up udp > logging on this 13th core and having the 12 instances of squid send the > log info over the internal network would work. > > Thought or better ideas? Problems with either of these scenarios? > > > Thanks in advance, > > // Joel > > jebrahimi@xxxxxxxxx > > Joel Ebrahimi > Solutions Engineer > Bivio Networks > 925.924.8681 > jebrahimi@xxxxxxxxx > >
